test: direct unit suite for the quantization conversion layer (63 tests)

[arham766]

What does this PR do?

Type of change: new tests

Part of the unit-coverage initiative in #1902 (one PR per module; this one is self-contained). Covers: quantize/restore mode layer: replace_quant_module insertion rules, register/unregister lifecycle, set_quantizer_by_cfg full-replacement vs partial-merge semantics across all three accepted formats (legacy formats pinned under pytest.warns(DeprecationWarning)), SequentialQuantizer creation/downgrade (guarded non-empty collection helpers), restore error directions (Unmatched vs Extra keys), cfg context-manager revert paths. Documents two real defects from the fused-experts name normalization (see issue).

Usage

N/A — tests only.

Testing

Suite is hermetic (CPU-only, no network, deterministic, <3s) and passes locally; the full collectable tests/unit tree passes with it (1999 passed). The suite was adversarially reviewed and verified to kill seeded mutations of the target module (details in #1902).

Before your PR is "Ready for review"

• Is this change backward compatible?: ✅
• If you copied code from any other sources or added a new PIP dependency, did you follow guidance in CONTRIBUTING.md: N/A
• Did you write any new necessary tests?: ✅
• Did you update Changelog?: N/A
• Did you get Claude approval on this PR?: N/A (external contributor)

Additional Information

Issue: #1902

Summary by CodeRabbit

• Tests
  • Added extensive unit-test coverage for quantization model conversion and restoration workflows.
  • Validated quantizer module replacement bookkeeping, including enable/disable defaults, in-place root conversion, and safeguards against already-quantized models.
  • Verified quantizer configuration application across multiple config formats (including wildcard matching and sequential quantizers), attribute update APIs (full/partial), and deprecated attribute handling with deprecation warnings.
  • Covered quantizer state extraction, metadata tracking, and strict round-trip restoration error cases.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: f00890b9-c654-4c87-babe-cfa11478fb22

📥 Commits

Reviewing files that changed from the base of the PR and between 7321acb and 2ebd0ff.

📒 Files selected for processing (1)

• tests/unit/torch/quantization/test_conversion.py

🚧 Files skipped from review as they are similar to previous changes (1)

• tests/unit/torch/quantization/test_conversion.py

---

📝 Walkthrough

Walkthrough

Adds a new unit test suite for quantization conversion, covering module replacement, registration, cfg-driven quantizer updates, attribute APIs, state restoration, conversion entrypoints, context-managed overrides, and a deprecated helper.

Changes

Conversion module test suite

Layer / File(s)	Summary
Test setup, helpers, and fixtures tests/unit/torch/quantization/test_conversion.py	Adds imports, shared constants, helper utilities, and dummy module classes used across the suite.
replace_quant_module and register/unregister lifecycle tests/unit/torch/quantization/test_conversion.py	Tests quantizer insertion behavior, parameter preservation, default states, rejection of already-quantized models, and register/unregister lifecycle with error handling.
set_quantizer_by_cfg behavior tests/unit/torch/quantization/test_conversion.py	Tests cfg application across new and legacy formats, deny-all/selective-enable patterns, parent-class scoping, sequential quantizer upgrade and downgrade, and enable-state broadcasting.
set_quantizer_attributes full and partial semantics tests/unit/torch/quantization/test_conversion.py	Tests full replacement and partial-merge semantics, callable and wildcard filters, sequential broadcast behavior, and validation of unsupported inputs.
Fused-experts quantizer name normalization tests/unit/torch/quantization/test_conversion.py	Tests normalization of fused-experts quantizer names for wildcard matching.
Quantizer state, metadata, and restoration tests/unit/torch/quantization/test_conversion.py	Tests quantizer_state coverage and equality, metadata recording, restore_quantizer_state roundtrips, buffer placeholder reconstruction, and NVFP4 promotion.
Convert/restore entrypoints tests/unit/torch/quantization/test_conversion.py	Tests convert_to_quantized_model/restore_quantized_model roundtrips, including in-place conversion and calibrated buffer presence.
Context overrides and deprecated API tests/unit/torch/quantization/test_conversion.py	Tests set_quantizer_by_cfg_context restoration behavior and deprecated set_quantizer_attribute warning and merge semantics.

Estimated code review effort: 3 (Moderate) | ~25 minutes

Possibly related issues

• [Coverage] Direct unit-test suites for five untested core modules (conversion, core_utils, tensor_quantizer, model_calib, export/postprocess) — plus bugs found along the way #1902: The new test file directly covers quantization conversion behaviors, matching this issue’s focus on conversion-module unit tests.

🚥 Pre-merge checks | ✅ 6

✅ Passed checks (6 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately summarizes the main change: adding a direct unit test suite for the quantization conversion layer.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Security Anti-Patterns	✅ Passed	Only a new test file was added, and it contains no torch.load/np.load/trust_remote_code/eval/exec/nosec patterns or new deps.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

🧹 Nitpick comments (1)

tests/unit/torch/quantization/test_conversion.py (1)

189-201: 🎯 Functional Correctness | 🔵 Trivial | ⚡ Quick win

Weak assertion for the "forward routes through quantizer" claim.

The comment on Line 199 states the forward call must route through the quantizer, but the test only checks that the call doesn't raise. By default TensorQuantizer has if_quant=True, if_calib=False and computes a dynamic amax per call, so a broken routing (quantizer silently bypassed) wouldn't necessarily raise here — the assertion doesn't actually pin the claimed behavior.

Consider a stronger check, e.g. patching/spying on input_quantizer.forward or comparing output against a manually-unquantized forward pass, to make this a real regression guard.

♻️ Example approach

-            model(torch.randn(2, 4))  # forward must route through the quantizer
+            called = []
+            orig_forward = model[1].input_quantizer.forward
+            model[1].input_quantizer.forward = lambda x: (called.append(True), orig_forward(x))[1]
+            model(torch.randn(2, 4))
+            assert called, "forward did not route through the quantizer"

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/unit/torch/quantization/test_conversion.py` around lines 189 - 201, The
forward-call assertion in test_register_replace_unregister_cycle is too weak to
prove routing through TensorQuantizer. Strengthen the check around
model(torch.randn(2, 4)) by spying on or patching
_QuantPlainAct.input_quantizer.forward, or by asserting the output differs from
a manually computed unquantized pass, so the test in
test_register_replace_unregister_cycle and the
replace_quant_module/register/unregister path actually verify quantizer usage
rather than just lack of exceptions.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@tests/unit/torch/quantization/test_conversion.py`:
- Around line 189-201: The forward-call assertion in
test_register_replace_unregister_cycle is too weak to prove routing through
TensorQuantizer. Strengthen the check around model(torch.randn(2, 4)) by spying
on or patching _QuantPlainAct.input_quantizer.forward, or by asserting the
output differs from a manually computed unquantized pass, so the test in
test_register_replace_unregister_cycle and the
replace_quant_module/register/unregister path actually verify quantizer usage
rather than just lack of exceptions.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 9ce486ed-389a-4f8c-8ac4-04ca3170e3f5

📥 Commits

Reviewing files that changed from the base of the PR and between b96a785 and 7321acb.

📒 Files selected for processing (1)

• tests/unit/torch/quantization/test_conversion.py

[arham766]

Consolidated into #1927 per maintainer feedback in #1902 — the suite was trimmed to only the lines codecov reports uncovered, with parametrization clusters deduplicated. Closing in favor of that PR.